Split a song into vocals and instrumental with Mel-Band RoFormer (MIT vocal weights in models/audio_separation). — ComfyUI node from LSDub. Run it in the…

Lip-sync for music videos in ComfyUI: dub the singers' mouths to the song, shot by shot, with QA. Every input is a slot (bring a clip and song, or generate them with LTX-2.5, Qwen-Image-Edit and ACE-Step); vocal split, lyric alignment, lip kits, sync QA.

LSDub

audio

model

precision

Split a song into vocals and instrumental with Mel-Band RoFormer (MIT vocal weights in models/audio_separation).

Name	Type	Default	Description
audio	AUDIO		The song mix (any sample rate; mono is doubled to stereo).
model	COMBO		Mel-Band RoFormer vocal checkpoint in models/audio_separation (MelBandRoformer_kim_vocals.ckpt).
precision	COMBO	fp32	fp32 is exact; bf16 is faster with a small quality loss. fp16 overflows in this network, so it is not offered.

Name	Type	Description
vocals	AUDIO	Vocals, stereo 44.1 kHz.
instrumental	AUDIO	Mix minus vocals, stereo 44.1 kHz.

LSDub Vocal Split (RoFormer)

Inputs (3)

Outputs (2)